Download the entire proceedings as an Adobe PDF - Eastern Snow ...

correspondence between simulated and observed spring Q suggests the model—to some degree—is
accounting for processes connecting the snowpack and spring river flow, e.g. sublimation, rainfall,
and soil infiltration.
To better understand the covariance between SWE and Q, alternate comparisons were made
using PWBM/ERA-40 monthly SWE and an estimate of when thaw is assumed to have occurred.
The month of thaw (TM,withTM− 1, and TM+ 1 indicating the month preceding and postceding
the thaw month, respectively) was determined with a step edge detection scheme applied to SSM/I
brightness temperatures (McDonald et al., 2004). Then, SWETM becomes monthly basin SWE
during TM (or TM− 1), and QTM is discharge in month TM. These alternate comparisons (across
all 179 basins) are defined (a) SWETM vs. spring Q, (b)SWETM-1 vs. spring Q, (c)SWETM-1 vs.
QTM+1, (d)SWETM-1 vs. QTM+1,2. R 2 s are highest for alternate comparison (b), which compared
SWE in the month before thaw (TM − 1) with spring (April–June) Q (Table 1). Yet, despite the
fact that the mean R 2 across western Eurasia improves from 0.15 (using default PWBM/ERA-40)
to 0.38 (alternate comparison b), little difference is noted with the remianing alternate comparisons
and other regions.
Lastly, we scaled spring Q using a factor S, whereS = PWBM monthly snow melt–runoff ratio,
with 0 < S < 1. Then, snowmelt Q each month is Qs =Q·S. Each occurrence of QS was then
summed resulting in a total QS each spring, for each basin. Using QS in place of the default Q (and
PWBM/ERA-40 SWE), we note a decrease in agreement across eastern Eurasia, with no change
across most of the domain. And although SWE from simulations with ERA-40, in general, explains
more than a third of the variation in Q, a large proportion of the interannual variability is not
due to SWE variability. When considering these results, it is interesting to note that Lammers et
al. (2006) recently found that annual simulated discharge across Alaska (drawn from three separate
models) was in poor agreement with observed discharge data between 1980–2001. Better agreements
across northwestern North America, eastern Eurasia (EE in Figure 3), and parts of western Eurasia
(WE) in this study are attributable to relatively higher snowfall rates and a greater interannual
variability in spring discharge (Figure 4b). Conversely, the region of eastern Eurasia with numerous
negative correlations is characterized by low spring discharge variability. Delays in snowmelt water
reaching river systems, which can be significant (Hinzman and Kane, 1991), are likely an additional
influence on these reported correlations. For large arctic basins, comparisons between snow storage
and discharge volume are complicated by the large temporal variation in basin thaw and the delays
in snowmelt water reaching the gauge. More meaningful comparisons between spatial SWE and
river discharge are possible through the use of hydrograph separation to partition discharge into
overland and baseflow components. This, however, requires the use of daily discharge data which
are more limited for the Pan-Arctic region.
CONCLUSIONS
In our comparisons of interannual variations in pre-melt SWE and spring Q, R 2 values are
highest (mean of 0.25 to 0.28 over all basins) when PWBM is driven by ERA-40, NNR or WM
climate data. Similar agreements are noted when SWE from the observed data analysis scheme are
used, which suggests that the hydrological model is capturing as much variability in the spring flow
as does the observed SWE scheme. Average R 2 determined from the SSM/I SWE and spring Q
comparisons are generally low, and a sizable majority (over 72%) of these correlations are negative.
The low variability and magnitude is likely related to saturation of the SSM/I algorithm at high SWE
129